A lossless audio codec based on hierarchical residual prediction

In this study, we propose a novel lossless audio codec that exhibits precise prediction performance leveraging a neural network with fast decoding speed. The proposed network structure is identical to the ResNet; we repeatedly predict on residual. Furthermore, we propose optimal parameter setting tactics for the recursive Golomb-Rice coding. Finally, we implement the codec and experimentally compare its performance with that of state-of-the-art codecs. We confirmed practical decoding speed and higher compression ability than other methods, except for Monkey's Audio.